Your customers are asking ChatGPT which product to buy. You can’t see those conversations, there is no Search Console for them, and if the model recommends a competitor, you lose the sale without ever knowing you were in the running.
That is the tracking problem in one sentence. The good news: you can measure most of it with a spreadsheet and 30 minutes a week, and the parts you can’t measure manually now have real tools.
This guide covers the three methods, in order of cost. If you want the background on why AI assistants cite some brands and ignore others, read our Generative Engine Optimization (GEO) guide first.
Method 1: manual prompt sampling (free)
The simplest tracker is a fixed list of prompts you run every week and a log of what comes back.
The mistake most teams make is sampling brand-name prompts. “Tell me about [your brand]” tells you almost nothing, because a buyer who already knows your name was going to find you anyway. Track the prompts where the model picks winners:
- Direct recommendations. “What’s the best AI writing tool for long-form SEO content?” No brand named. Who does the model volunteer?
- Comparisons. “Compare [you] vs [competitor] for [use case].” Does the model represent your features accurately, and which way does it lean?
- Problem prompts. “My blog traffic dropped after a Google update, what should I do?” These reveal whether the model associates your brand with the problems you solve.
Build 15-25 prompts across those three groups, then run them on a schedule. Two rules keep the data honest:
- Use a fresh or logged-out session with memory off. ChatGPT personalizes answers, and your own history will skew results toward brands you already research.
- Log the same fields every week: mentioned yes/no, position in the answer (first pick, listed, footnote), sentiment, and which competitors appeared. A plain spreadsheet works.
Run the set in ChatGPT, then repeat in Perplexity and Gemini. The three differ more than you’d expect: Perplexity cites sources aggressively and leans on forums and recent web pages, Gemini tracks Google’s live index closely, and ChatGPT blends model memory with optional web search. A brand can be the first pick in one and absent from another.
Method 2: your own analytics
You already have partial visibility data and probably haven’t looked at it.
When ChatGPT cites your page and a user clicks through, the visit arrives with a chatgpt.com referrer. In GA4, build a report filtered on session source containing chatgpt.com, perplexity.ai, and gemini.google.com. That is real, attributable AI-assistant traffic, and its trend line tells you whether your visibility work is moving.
Two caveats. First, this only counts clicks, and many AI answers satisfy the user without one, so treat it as a floor, not a total. Second, Google Search Console does not separate AI Overview impressions from regular results, so your GSC numbers already blend the two.
Method 3: dedicated tracking tools
Manual sampling stops scaling somewhere around 25 prompts or 2 competitors, whichever comes first. The paid trackers all automate the same loop: run a prompt set on a schedule, parse the answers, and report mentions, position, and share of voice over time.
Tools we’d shortlist, based on what each vendor documents:
| Tool | What it does | Standout |
|---|---|---|
| Semrush AI visibility checker | Free snapshot of your brand’s presence in AI search | Free, fastest first look |
| SE Ranking ChatGPT Visibility Tracker | Scans ChatGPT answers for your target keywords, reports which mention your brand | Keyword-first workflow, familiar if you already rank-track |
| Otterly | Monitors brand appearances across ChatGPT, Perplexity, and Google AI Overviews, with competitor comparison | Multi-engine coverage in one dashboard |
| Siftly | Runs your customer prompts daily across major AI engines, flags mentions, sentiment, and hallucinations | Hallucination flagging |
| BrandPeek | Brand-visibility monitoring plus an AI Overview analyzer and AI fact-sheet generator | Free premium access for Agility Writer subscribers |
Full disclosure on that last row: BrandPeek is our partner product. Agility Writer subscribers unlock its premium features without a separate subscription by connecting their account, and you can transfer unused Agility Writer credits to it. That arrangement is why we know it best, and you should weigh our recommendation accordingly.
The weekly loop that makes tracking useful
Data you don’t act on is a hobby. The loop that pays for itself:
- Run your prompt set (manually or via a tool) on the same weekday.
- Diff against last week. New mentions, lost mentions, position changes, new competitors in answers.
- For every prompt where a competitor beats you, read the sources the AI cited. Those pages are your content gap list.
- Publish or update one page per gap. Comparison pages work particularly well here, because “X vs Y” prompts are exactly what buyers ask assistants. Agility Writer’s Fact Sheets keep the product details consistent across those pages, and the Comparison Page Ideas Generator builds the target list.
- Re-check the prompt in two to four weeks. AI answers move faster than blue links, so wins show up sooner than you’re used to.
Expect drift. An answer that features you on Monday can feature a competitor by Friday after a silent model update. That volatility is exactly why a weekly baseline beats a one-off audit.
What to do when you’re not cited
Tracking tells you where you’re invisible. Fixing it is content work: pages that answer the prompt directly, structured so a model can lift the answer, with your brand and its facts stated plainly. Question-style headings and self-contained sections help, which is what the AI-Friendly Outline option in Agility Writer’s writing modes produces, and the GEO guide covers the full playbook.
To try that workflow on a real article, start the $1 trial.